06.03.2026 19:49
👍 1
🔁 0
💬 0
📌 0
"it's still not neutral"
i never said it was!!!
the weights don't age out of compatibility or compliance.
they are not. open models from disparate sources that you can run on your own hardware do not bake in authoritarianism - something I explained pretty clearly already.
you aren't arguing with my actual words but instead inventing a phantom me who claimed that all technology is politically neutral - something i literally never claimed.
i even *strengthened* your own badly made argument and then argued against that but you won't stop strawmanning me
you are literally posting a screenshot of me quoting you and *agreeing* with the statement that tech is not politically neutral.
my brother in christ the only person in this thread talking about tech being politically neutral is you
as for "needs to be updated" - no? it just does what it does. you might want a newer model eventually but nobody is forcing you.
is your hammer controlled by the hammer manufacturer because one day you might buy a better hammer?
as for your last point... yes? that's the point? you keep saying "technology is not politically neutral" but nobody has claimed it *is* neutral?
but as you say, "it might turn out socially beneficial in the long run". in other words - the political outcome is not determined or inherent in the tech!
your Linux take is very, very, dated (Linux is actually easier to install and use than Windows these days)
but that isn't even really important as the analogy doesn't hold water - running a local LLM is literally "install LM Studio or Ollama, download model, done."
there is *so much* competition in the frontier open-weight space. meta, mistral, alibaba (qwen), deepseek, cohere, google - to name a few. these are actors with completely divergent geopolitical and commercial interest.
once you download them, nobody can revoke your access or push an update that changes how they work. they'll work exactly the same, forever. model weights they don't phone home or expire.
it's more like... having a copy of photoshop on a floppy disk in 1995 than a creative cloud subscription.
i don't think you understand how local models work? you're pattern-matching "AI" onto a "big tech SaaS" mental model. so the adobe/microsoft comparison is just... wrong? like, factually wrong.
if you download Llama weights from huggingface, you *have them*. they're yours.
3. the gap between frontier and locally-runnable has been shrinking! you can automate your workflow & build really good tools on top of them, running on a single consumer GPU (or even a phone!).
2. all AI labs are competing with each other. some of them release open weight models (Meta, Mistral, various Chinese labs) for various reasons. this is not a single actor controlling the pipeline, it's a bunch of players with divergent interests, so nobody is dictating terms.
but that argument is getting weaker as time goes on, not stronger.
1. LoRA, QLoRA, full fine-tuning techniques can be done on consumer hardware today to reshape a base model's behaviour, knowledge, etc. to get something completely custom.
to steelman what I think you're *trying* to say:
it's true that the training pipeline controls what capabilities exist in the first place, and this is still a fairly centralised capital-intensive process. the frontier is shaped by a few actors, local inference is just a hand-me-down from power.
you may as well be stating that widespread literacy is equivalent to guns because states used literacy for propaganda.
"technology is not politically neutral" sure, nothing is. but a tech's political leanings are not *inherent and fixed* by whoever built it.
technologies have uses. your gun example is... bad. a gun has one use - projecting lethal force. an LLM is a general purpose knowledge tool with huge use-space
i can't build a printing press or a smartphone either but those inventions still expanded access to literacy, knowledge, banking, etc.
long reply:
you're conflating the ability to create a technology from scratch with the ability to wield it as a means of production.
and actually yes, being able to run your own copy of software is actually incredible? free open-source software is amazing for democratisation and decentralisation!
Accountability tracker for the UK Government's Warm Homes: Social Housing Fund scheme
Pulls the official data tables from the government website on a monthly basis
I'll try to correct any issues found as I'm not a domain expert
bsky.app/profile/justin
You can run your own AI models on your own hardware and they're good and they're getting better and the hardware is not even that expensive.
the towers were here long before we were and they'll likely be here long after we're gone. you learn to live with the occasional headache and the constant background hum. but you never get used to the dreams.
youtube you can keep changing your internal api but it won't stop me
just updated the plugin to address some issues (and switch to the atcute packages) and add a couple of QoL things
the viewer SPA now fully supports social cards when deployed to cloudflare pages!
openai's actual contract language is out. 'all lawful purposes' + EO 12333, which is how the NSA captures US persons' data from international tapping points. the 'prohibition on domestic surveillance' was in the press release. not in the contract.
truly a stars align moment
woah wtf I had no idea this was created literally a week ago